If you sell something online, you have already noticed that the places people trust are not the places you can buy an ad in. A question typed into ChatGPT or Perplexity is answered from Reddit threads, forum comparisons and reviews. A question typed into Google increasingly gets an AI answer built from the same material. Your ad is not in it. Somebody's opinion is.
So the practical question for a marketer in 2026 is how to get real people to talk about your product, and how to do that without buying fake reviews, which is now illegal in the two largest English-speaking markets. This guide answers both, and ends with something we can vouch for: what three weeks of doing it the honest way did for this site.
Why mentions, not links, decide what AI says about you
Profound analysed 680 million citations from AI answers between August 2024 and June 2025 and published the shares by platform. Among Perplexity's ten most cited sources, Reddit alone accounts for 46.7%. In Google's AI Overviews it is 21%. ChatGPT leans on Wikipedia instead, at 47.9% of its top ten. The pattern is the same in every case: the model quotes places where people compare things, not places where brands describe themselves.
That is the whole reason brand mentions became a line in marketing budgets. A mention in a thread that the model already trusts does what a backlink did ten years ago. The difference is that a backlink could be bought quietly, and a mention that is bought quietly is now a regulatory problem.
What the law says now
Three sets of rules matter, and they agree with each other.
United States. The FTC's rule on consumer reviews and testimonials was announced on 14 August 2024 and took effect 60 days after publication in the Federal Register, on 21 October 2024. It bans six things: fake or false reviews including AI-generated ones, paying for a review conditioned on its sentiment, insider reviews without disclosure, company-controlled review sites posing as independent, suppressing negative reviews by threat, and buying fake followers or views. The rule lets the FTC seek civil penalties against knowing violators. Note what it does not ban: paying somebody to write an honest review, as long as the payment does not depend on the review being positive and the relationship is disclosed.
United Kingdom. The Digital Markets, Competition and Consumers Act made fake reviews and concealed incentivised reviews a banned practice from 6 April 2025. The CMA's guidance CMA208, dated 4 April 2025, defines a concealed incentivised review in two lines: a person has been commissioned to provide the review, and that fact is not made apparent. Commissioning includes money, commissions, discounts, vouchers and free loans. Missing, obscure or hidden disclosure counts as concealment. The same document treats offering services to procure such reviews as a banned practice in its own right, which is why we run our moderation the way we do.
Reddit. The platform-wide rules are shorter than the law and say the same thing. Rule 2: participate authentically in communities where you have a personal interest, do not spam or engage in content manipulation. Rule 5: be authentic, do not intentionally mislead others or impersonate an individual or entity. Subreddit moderators add their own limits on self-promotion, and they enforce them faster than any regulator.
Put together: you may pay a person to write. You may not pay for the opinion, and you may not hide the payment. Everything in the rest of this guide follows from that.
What gets rejected, and why it should
A brief that survives the rules above still fails if the writing reads like an ad, because the community will remove it before any regulator sees it. Our editorial moderation for tasks on AntiMedia's author marketplace rejects submissions for a short list of reasons, and the list is worth knowing whether or not you use the marketplace:
- No disclosure where the platform requires one. A post that hides the relationship is a concealed incentive, full stop.
- Copied or generated text. If the same paragraph exists anywhere else, or reads like a template, it goes back. Communities and AI systems both filter it out, so it earns nothing.
- No first-hand experience. "Great service, recommend" is a fake review by the CMA's definition, because it purports to be based on experience it does not have.
- A claim the author cannot back. Bonus amounts, win rates, "best in the market". These are the sentences that get a brand reported.
- The wrong place. Posting a product mention in a subreddit whose rules forbid it, or replying to a three-year-old thread nobody reads.
The useful mental test for any brief: would the author be comfortable if the moderator, the platform and the FTC all read it? If the answer needs a second thought, rewrite the brief.
A brief that works, and one that does not
Here is the kind of task that produces mentions worth having. Both are for the same imaginary betting exchange.
Bad brief. "Write 5 positive reviews about BetExample on Reddit and forums. Mention the 200% welcome bonus and fast withdrawals. Do not mention that this is sponsored."
Everything in it is a violation: the number of reviews, the fixed sentiment, the bonus claim, the concealment.
Good brief. "You have used BetExample for at least two weeks. In a thread where people ask how exchanges differ from bookmakers, explain in your own words what was better and what was worse than the two other services you used, including one thing you disliked. Disclose that BetExample paid for your time, in the form the subreddit's rules require. Do not quote bonus amounts. Link only if the thread's rules allow links."
The second brief costs more per placement, because it needs a real user with a real history. It also produces the only kind of text an AI answer will ever quote: specific, comparative, and signed by a person.
What honest mentions cost
There is no market price, so here is how to think about it instead.
A placement has three costs: the author's time, the moderation that checks the text before it goes live, and the risk the author carries on their own account. Cheap offers skip the second and hide the third. Compare that with what you are paying elsewhere. In our analysis of fake store pages we quoted Netcraft's figure of $50 to $350 in affiliate payouts per depositing player in gambling. One honest thread that sends two people who stay is worth more than a hundred that get removed, and it is still there next year.
Two rules for budgeting. Pay for the writing and the disclosure, never for the sentiment: the moment the payment depends on a positive tone you are inside the FTC's rule. And measure mentions the way you measure any channel: registrations and retention from the referrer, not the number of posts.
Three weeks of honest content: what it did for us
Everything above is advice. This part is a receipt.
This site published its first article on 15 August 2026. Since then it has run under the rules we ask of authors: no bought links, no ads, no rewritten press releases, every dated claim sourced, and at least a third of every piece made of something that does not exist elsewhere. Here is what Google and the AI engines did with that by 8 September 2026, straight from Search Console and from a test we ran ourselves.
| Measure, 10 August to 6 September 2026 | Value |
|---|---|
| Pages indexed by Google | 31 |
| Search impressions | 428 |
| Search clicks | 29 |
| Average click-through rate | 6.8% |
| Average position | 25.5 |
| Impressions on the brand query "antimedia" | 85, with 8 clicks |
| Impressions in Google's generative AI features (AI Overviews and AI Mode) | 31 |
| Google Discover impressions | 0 |
The generative AI line is the one advertisers ask about. Search Console now has a separate beta report for it, and it says our pages were shown inside Google's AI answers 31 times in 28 days, mostly the homepage. That is not zero, and it is not a channel.
Then we tested it from the other side. On 8 September we took seven questions that our articles answer directly, on trackers, postbacks, Android developer verification, Google Discover, iGaming SEO, PWA services and Telegram channels, and asked each of them in Google's AI Mode and in Perplexity. Fourteen answers. Google AI Mode built them from 4 to 18 domains each, Perplexity from 15 sources each. Our site was cited in none of them. The domains that were cited were tool vendors' own blogs, Medium posts, affiliate media, developer documentation and Reddit.
We are publishing that number because it is the honest baseline, and because it makes the point better than a success story would. Three weeks of original writing on a new domain gets you indexed, gets you brand-name searches and gets you a first appearance in Google's AI features. It does not get you quoted by AI answers, because those answers are assembled from places where other people talk about you. The way in is not a better article on your own site. It is a real person in a real thread saying what they found, with the disclosure the law now requires. That is the service we built, and this is why.
How to start, in order
- Write the disclosure line first, per platform, and put it in the brief.
- Pick the questions your support team hears most. Those are the threads that exist already.
- Find authors who have used the product, or give them time to. Two weeks is the minimum for a review that is not fake by definition.
- Brief for comparison, not praise. Ask for one thing they disliked.
- Have somebody who is not the author read every text against the list above before it goes live.
- Measure registrations and 30-day retention by referrer. Cut what does not hold.
Frequently asked questions
Is it legal to pay people to mention my brand? Yes, in the US and the UK, if the payment does not depend on the opinion expressed and the relationship is disclosed. Paying for a positive review, or hiding the payment, is banned in both.
Does a disclosed mention still help with AI visibility? The disclosure does not stop a model from reading the text. What matters is that the text is specific and sits in a place the model already cites, which for most questions means Reddit and specialist forums.
How many mentions do I need? Fewer than any seller will tell you. One thread that people keep finding is worth more than fifty that get removed. Start with the questions your customers actually ask.
Can I use AI to write the mentions? The FTC rule names AI-generated reviews explicitly as fake when they misrepresent a consumer's experience, and communities remove them on sight. Use it to draft the brief, not the post.
How long until it shows up in AI answers? Our own site got into Google's generative AI report inside a month but was cited in none of fourteen test answers after three weeks. Plan in quarters, and put the money into places other people write, not only into your own pages.
Method note
Search figures are from Google Search Console for the property antimedia.co, 28 days ending 6 September 2026, web search type, plus the Generative AI features beta report and the Discover report for the same window. The citation test was run on 8 September 2026 from a desktop browser in English: seven questions, each asked once in Google AI Mode and once in Perplexity, checking every linked source in the answer for our domain. ChatGPT was not included because it requires an account for search. No platform figures from the author marketplace appear in this article.